Project Info
This project did not submit a demo video on Devpost.
Inspiration
63 million Americans provide unpaid care to family members, primarily aging parents and children that require special assistance. A complex ecosystem of programs that pay our caregivers exists — from In-Home Supportive Services (IHSS), Paid Family Leave (PFL), to VA benefits. However, caregivers face many systemic obstacles accessing them: (1) a large, interconnected set of rules for eligibility, making it difficult to understand what you actually qualify for. These rules also vary by state (2) confusing application requirements: different programs require different sets of forms, all rich with jargon that's hard to understand. While existing companies like Givers.com and Wellthy are shallow, low-tech wrappers around human agencies specializing in a handful of programs in a few states, AI can be a great tool to help navigate caretakers through this process with far greater program scope and lower costs.
What it does
(1) Intuitive Intake form: basic assessment of caregiver situation, financial status, medical needs of care recipient. Routing agent calls on specialist agents based on user context (2) Multi-agent application assistance and optimization: specialist agents are created and trained with RAG on vector databases consisting of hundreds of pages of eligibility information, online guides, and official government documents. In a shared agent space, they discuss likelihood of eligibility and cross-program interactions. They resolve conflicts and decide on an application strategy and prioritization system that maximizes the user's chances of getting benefits (3) The dashboard surfaces all of these potentially eligible programs along with status and notes. When clicked, the respective specialist agent pulls up all the forms necessary to apply and determines what remaining information is needed, breaking it into easy short-answer questions. The user answers the questions, only having to interact with the completed forms at the end to check over them, and easily exports them into one complete PDF to turn in in-person (applications are usually filed in-person at government offices.) (4) Care Calendar & Agentic Assistant: a centralized calendar for caregivers to see doctor's appointments, social workers' visits, etc. An agent scans emails and calendars, automatically suggests events to add, and sends text reminders. (5) Timekeeping: many of these benefits are also lost out because of missed renewals. Caregivers need to maintain timesheets of care given and apply for benefits renewal yearly. Ilera provides both an easy interface to log caregiving as well as a journal feature. Concerning journal features automatically prompt calendar suggestions (i.e: a medical incident creates a suggested doctor's appointment in the calendar) (6) Document Store for important records, auto-populated with complete applications
How we built it
NextJS TypeScript for frontend FastAPI on backend with an MCP server that calls Poke and prompts it to add events to the calendar. I created a Poke recipe integrating with Outlook, GMail + their calendars. With one click during the signup process, users sign up for this integration. Redis for vector database containing official program documents, online eligibility guides Band + Anthropic API to create a shared agent space with Claude-based agents specializing in different program families Devin Enterprise for coding!
Challenges we ran into
I had some trouble with the Redis vector database. It kept drifting back to 5 chunks due to branch issues. Finding the correct documents and setting up the vector database meant doing a lot of domain research.
Accomplishments we're proud of
Setting up my first MCP server! Successfully getting Poke to suggest events for the Calendar. Orchestrating a complex application workflow (intake form -> multi-agent eligibility determination -> determining correct forms to complete -> deciding what information is needed -> completing and stitching the forms on the backend)
What we learned
AI is much more capable of interacting with users outside of applications than I thought was possible! I believe the most successful version of this type of product is one that can use tech to synthesize information and action plans across many fragmented sources and data streams on the users' end.
What's next
There are many opportunities to extend the agentic part of the app using tools like Browserbase: automatically submitting timekeeping entries as IHSS timesheet-formatted PDFs to the online portal syncing with EHR systems and adding real functionality to the "Create Appointment" button adding the new Apple Messages integration with Poke Additionally, expanding the document store from a stub feature would make this app more useful.
This project did not link a GitHub repository.
Analysis
No indexed repository for this project, so there are no commit stats to show.
Technology
- Next.jsUnchecked
- PythonUnchecked
- RedisUnchecked
No repository was indexed for this project, so these Devpost claims have not been checked against code.
AI coding agents
No repository was indexed, so agent usage could not be checked.
Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.
Codebase size
No repository was indexed, so there is no codebase to measure.
This project did not link a GitHub repository, so there is nothing to diagram.
This project did not link a GitHub repository, so its feature claims have not been checked against code.
Export this project's context (description, README, evidence, key source files) to chat with an AI agent elsewhere.